PURPOSE: To provide a neural network learning system capable of efficiently and rapidly executing learning while suppressing the increment of hardwares by using plural neural networks having respectively different numbers of intermediate layer units in respect to a neural network learning system having wide application fields such as the recognition of patterns and characters, various control, and so on.

CONSTITUTION: The neural network system arraying plural hierarchical neural networks each of which is constituted of an input layer 20, one intermediate layer 21 or more and one or more output layers 22 is provided with the common input layer 20 corresponding to two or more in plural neural networks or the common input layer 20 corresponding to two neural networks or more, one intermediate layer 21 or more and a learning control means 25 for controlling the learning of plural neural networks.